On Mon, Jul 12, 2021 at 01:14:24PM +0200, Christoph Hellwig wrote:
> Disabling quota accounting is hairy, racy code with all kinds of pitfalls.
> And it has a very strange mind set, as quota accounting (unlike
> enforcement) really is a propery of the on-disk format. There is no good
> use case for supporting this.
>
> Signed-off-by: Christoph Hellwig <hch@xxxxxx>
> - if ((dqtype & XFS_QMOPT_UQUOTA) && q->qi_uquotaip) {
> - xfs_irele(q->qi_uquotaip);
> - q->qi_uquotaip = NULL;
> - }
> - if ((dqtype & XFS_QMOPT_GQUOTA) && q->qi_gquotaip) {
> - xfs_irele(q->qi_gquotaip);
> - q->qi_gquotaip = NULL;
> - }
> - if ((dqtype & XFS_QMOPT_PQUOTA) && q->qi_pquotaip) {
> - xfs_irele(q->qi_pquotaip);
> - q->qi_pquotaip = NULL;
> - }
> + mutex_lock(&mp->m_quotainfo->qi_quotaofflock);
> + mp->m_qflags &= ~(flags & XFS_ALL_QUOTA_ENFD);
> + spin_lock(&mp->m_sb_lock);
> + mp->m_sb.sb_qflags = mp->m_qflags;
> + spin_unlock(&mp->m_sb_lock);
> + mutex_unlock(&mp->m_quotainfo->qi_quotaofflock);
>
> -out_unlock:
> - if (error && qoffstart)
> - xfs_qm_qoff_logitem_relse(qoffstart);
> - mutex_unlock(&q->qi_quotaofflock);
> - return error;
> + /* XXX what to do if error ? Revert back to old vals incore ? */
> + return xfs_sync_sb(mp, false);
May be too strict, but I wonder if we shouldn't shut the FS down in case we fail
here?
Patch looks good though.
Reviewed-by: Carlos Maiolino <cmaiolino@xxxxxxxxxx>
--
Carlos
